Your First AI application¶
Going forward, AI algorithms will be incorporated into more and more everyday applications. For example, you might want to include an image classifier in a smart phone app. To do this, you'd use a deep learning model trained on hundreds of thousands of images as part of the overall application architecture. A large part of software development in the future will be using these types of models as common parts of applications.
In this project, you'll train an image classifier to recognize different species of flowers. You can imagine using something like this in a phone app that tells you the name of the flower your camera is looking at. In practice you'd train this classifier, then export it for use in your application. We'll be using this dataset from Oxford of 102 flower categories, you can see a few examples below.
The project is broken down into multiple steps:
- Load the image dataset and create a pipeline.
- Build and Train an image classifier on this dataset.
- Use your trained model to perform inference on flower images.
We'll lead you through each part which you'll implement in Python.
When you've completed this project, you'll have an application that can be trained on any set of labeled images. Here your network will be learning about flowers and end up as a command line application. But, what you do with your new skills depends on your imagination and effort in building a dataset. For example, imagine an app where you take a picture of a car, it tells you what the make and model is, then looks up information about it. Go build your own dataset and make something new.
Import Resources¶
I downgraded tensorflow to 2.14.0 because the curret version 2.18.0 has dependenciy issues with keras
In [1]:
!pip install tensorflow==2.14.0 # Downgrade to guarantee Keras 2.x
import tensorflow as tf
keras = tf.keras # Safe to use now
Collecting tensorflow==2.14.0 Downloading tensorflow-2.14.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (4.1 kB) Requirement already satisfied: absl-py>=1.0.0 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (1.4.0) Requirement already satisfied: astunparse>=1.6.0 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (1.6.3) Requirement already satisfied: flatbuffers>=23.5.26 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (25.2.10) Requirement already satisfied: gast!=0.5.0,!=0.5.1,!=0.5.2,>=0.2.1 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (0.6.0) Requirement already satisfied: google-pasta>=0.1.1 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (0.2.0) Requirement already satisfied: h5py>=2.9.0 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (3.12.1) Requirement already satisfied: libclang>=13.0.0 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (18.1.1) Collecting ml-dtypes==0.2.0 (from tensorflow==2.14.0) Downloading ml_dtypes-0.2.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (20 kB) Requirement already satisfied: numpy>=1.23.5 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (1.26.4) Requirement already satisfied: opt-einsum>=2.3.2 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (3.4.0) Requirement already satisfied: packaging in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (24.2) Requirement already satisfied: protobuf!=4.21.0,!=4.21.1,!=4.21.2,!=4.21.3,!=4.21.4,!=4.21.5,<5.0.0dev,>=3.20.3 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (3.20.3) Requirement already satisfied: setuptools in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (75.1.0) Requirement already satisfied: six>=1.12.0 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (1.17.0) Requirement already satisfied: termcolor>=1.1.0 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (2.5.0) Requirement already satisfied: typing-extensions>=3.6.6 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (4.13.1) Collecting wrapt<1.15,>=1.11.0 (from tensorflow==2.14.0) Downloading wrapt-1.14.1-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (6.7 kB) Requirement already satisfied: tensorflow-io-gcs-filesystem>=0.23.1 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (0.37.1) Requirement already satisfied: grpcio<2.0,>=1.24.3 in /usr/local/lib/python3.11/dist-packages (from tensorflow==2.14.0) (1.70.0) Collecting tensorboard<2.15,>=2.14 (from tensorflow==2.14.0) Downloading tensorboard-2.14.1-py3-none-any.whl.metadata (1.7 kB) Collecting tensorflow-estimator<2.15,>=2.14.0 (from tensorflow==2.14.0) Downloading tensorflow_estimator-2.14.0-py2.py3-none-any.whl.metadata (1.3 kB) Collecting keras<2.15,>=2.14.0 (from tensorflow==2.14.0) Downloading keras-2.14.0-py3-none-any.whl.metadata (2.4 kB) Requirement already satisfied: wheel<1.0,>=0.23.0 in /usr/local/lib/python3.11/dist-packages (from astunparse>=1.6.0->tensorflow==2.14.0) (0.45.1) Requirement already satisfied: mkl_fft in /usr/local/lib/python3.11/dist-packages (from numpy>=1.23.5->tensorflow==2.14.0) (1.3.8) Requirement already satisfied: mkl_random in /usr/local/lib/python3.11/dist-packages (from numpy>=1.23.5->tensorflow==2.14.0) (1.2.4) Requirement already satisfied: mkl_umath in /usr/local/lib/python3.11/dist-packages (from numpy>=1.23.5->tensorflow==2.14.0) (0.1.1) Requirement already satisfied: mkl in /usr/local/lib/python3.11/dist-packages (from numpy>=1.23.5->tensorflow==2.14.0) (2025.1.0) Requirement already satisfied: tbb4py in /usr/local/lib/python3.11/dist-packages (from numpy>=1.23.5->tensorflow==2.14.0) (2022.1.0) Requirement already satisfied: mkl-service in /usr/local/lib/python3.11/dist-packages (from numpy>=1.23.5->tensorflow==2.14.0) (2.4.1) Requirement already satisfied: google-auth<3,>=1.6.3 in /usr/local/lib/python3.11/dist-packages (from tensorboard<2.15,>=2.14->tensorflow==2.14.0) (2.27.0) Collecting google-auth-oauthlib<1.1,>=0.5 (from tensorboard<2.15,>=2.14->tensorflow==2.14.0) Downloading google_auth_oauthlib-1.0.0-py2.py3-none-any.whl.metadata (2.7 kB) Requirement already satisfied: markdown>=2.6.8 in /usr/local/lib/python3.11/dist-packages (from tensorboard<2.15,>=2.14->tensorflow==2.14.0) (3.7) Requirement already satisfied: requests<3,>=2.21.0 in /usr/local/lib/python3.11/dist-packages (from tensorboard<2.15,>=2.14->tensorflow==2.14.0) (2.32.3) Requirement already satisfied: tensorboard-data-server<0.8.0,>=0.7.0 in /usr/local/lib/python3.11/dist-packages (from tensorboard<2.15,>=2.14->tensorflow==2.14.0) (0.7.2) Requirement already satisfied: werkzeug>=1.0.1 in /usr/local/lib/python3.11/dist-packages (from tensorboard<2.15,>=2.14->tensorflow==2.14.0) (3.1.3) Requirement already satisfied: cachetools<6.0,>=2.0.0 in /usr/local/lib/python3.11/dist-packages (from google-auth<3,>=1.6.3->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (5.5.2) Requirement already satisfied: pyasn1-modules>=0.2.1 in /usr/local/lib/python3.11/dist-packages (from google-auth<3,>=1.6.3->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (0.4.1) Requirement already satisfied: rsa<5,>=3.1.4 in /usr/local/lib/python3.11/dist-packages (from google-auth<3,>=1.6.3->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (4.9) Requirement already satisfied: requests-oauthlib>=0.7.0 in /usr/local/lib/python3.11/dist-packages (from google-auth-oauthlib<1.1,>=0.5->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (2.0.0) Requirement already satisfied: charset-normalizer<4,>=2 in /usr/local/lib/python3.11/dist-packages (from requests<3,>=2.21.0->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (3.4.1) Requirement already satisfied: idna<4,>=2.5 in /usr/local/lib/python3.11/dist-packages (from requests<3,>=2.21.0->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (3.10) Requirement already satisfied: urllib3<3,>=1.21.1 in /usr/local/lib/python3.11/dist-packages (from requests<3,>=2.21.0->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (2.3.0) Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.11/dist-packages (from requests<3,>=2.21.0->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (2025.1.31) Requirement already satisfied: MarkupSafe>=2.1.1 in /usr/local/lib/python3.11/dist-packages (from werkzeug>=1.0.1->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (3.0.2) Requirement already satisfied: intel-openmp<2026,>=2024 in /usr/local/lib/python3.11/dist-packages (from mkl->numpy>=1.23.5->tensorflow==2.14.0) (2024.2.0) Requirement already satisfied: tbb==2022.* in /usr/local/lib/python3.11/dist-packages (from mkl->numpy>=1.23.5->tensorflow==2.14.0) (2022.1.0) Requirement already satisfied: tcmlib==1.* in /usr/local/lib/python3.11/dist-packages (from tbb==2022.*->mkl->numpy>=1.23.5->tensorflow==2.14.0) (1.2.0) Requirement already satisfied: intel-cmplr-lib-rt in /usr/local/lib/python3.11/dist-packages (from mkl_umath->numpy>=1.23.5->tensorflow==2.14.0) (2024.2.0) Requirement already satisfied: intel-cmplr-lib-ur==2024.2.0 in /usr/local/lib/python3.11/dist-packages (from intel-openmp<2026,>=2024->mkl->numpy>=1.23.5->tensorflow==2.14.0) (2024.2.0) Requirement already satisfied: pyasn1<0.7.0,>=0.4.6 in /usr/local/lib/python3.11/dist-packages (from pyasn1-modules>=0.2.1->google-auth<3,>=1.6.3->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (0.6.1) Requirement already satisfied: oauthlib>=3.0.0 in /usr/local/lib/python3.11/dist-packages (from requests-oauthlib>=0.7.0->google-auth-oauthlib<1.1,>=0.5->tensorboard<2.15,>=2.14->tensorflow==2.14.0) (3.2.2) Downloading tensorflow-2.14.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (489.9 MB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 489.9/489.9 MB 3.4 MB/s eta 0:00:000:00:0100:01 Downloading ml_dtypes-0.2.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (1.0 MB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.0/1.0 MB 49.5 MB/s eta 0:00:00 Downloading keras-2.14.0-py3-none-any.whl (1.7 MB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.7/1.7 MB 68.3 MB/s eta 0:00:00 Downloading tensorboard-2.14.1-py3-none-any.whl (5.5 MB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 5.5/5.5 MB 104.3 MB/s eta 0:00:0000:01 Downloading tensorflow_estimator-2.14.0-py2.py3-none-any.whl (440 kB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 440.7/440.7 kB 28.3 MB/s eta 0:00:00 Downloading wrapt-1.14.1-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl (78 kB) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 78.4/78.4 kB 6.2 MB/s eta 0:00:00 Downloading google_auth_oauthlib-1.0.0-py2.py3-none-any.whl (18 kB) Installing collected packages: wrapt, tensorflow-estimator, keras, google-auth-oauthlib, tensorboard, ml-dtypes, tensorflow Attempting uninstall: wrapt Found existing installation: wrapt 1.17.2 Uninstalling wrapt-1.17.2: Successfully uninstalled wrapt-1.17.2 Attempting uninstall: keras Found existing installation: keras 3.5.0 Uninstalling keras-3.5.0: Successfully uninstalled keras-3.5.0 Attempting uninstall: google-auth-oauthlib Found existing installation: google-auth-oauthlib 1.2.1 Uninstalling google-auth-oauthlib-1.2.1: Successfully uninstalled google-auth-oauthlib-1.2.1 Attempting uninstall: tensorboard Found existing installation: tensorboard 2.18.0 Uninstalling tensorboard-2.18.0: Successfully uninstalled tensorboard-2.18.0 Attempting uninstall: ml-dtypes Found existing installation: ml-dtypes 0.4.1 Uninstalling ml-dtypes-0.4.1: Successfully uninstalled ml-dtypes-0.4.1 Attempting uninstall: tensorflow Found existing installation: tensorflow 2.18.0 Uninstalling tensorflow-2.18.0: Successfully uninstalled tensorflow-2.18.0 ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. tensorflow-decision-forests 1.11.0 requires tensorflow==2.18.0, but you have tensorflow 2.14.0 which is incompatible. dopamine-rl 4.1.2 requires gymnasium>=1.0.0, but you have gymnasium 0.29.0 which is incompatible. pandas-gbq 0.26.1 requires google-api-core<3.0.0dev,>=2.10.2, but you have google-api-core 1.34.1 which is incompatible. gcsfs 2024.10.0 requires fsspec==2024.10.0, but you have fsspec 2025.3.2 which is incompatible. bigframes 1.36.0 requires rich<14,>=12.4.4, but you have rich 14.0.0 which is incompatible. tf-keras 2.18.0 requires tensorflow<2.19,>=2.18, but you have tensorflow 2.14.0 which is incompatible. tensorflow-text 2.18.1 requires tensorflow<2.19,>=2.18.0, but you have tensorflow 2.14.0 which is incompatible. tensorstore 0.1.71 requires ml_dtypes>=0.3.1, but you have ml-dtypes 0.2.0 which is incompatible. Successfully installed google-auth-oauthlib-1.0.0 keras-2.14.0 ml-dtypes-0.2.0 tensorboard-2.14.1 tensorflow-2.14.0 tensorflow-estimator-2.14.0 wrapt-1.14.1
2025-04-18 04:30:24.254742: E tensorflow/compiler/xla/stream_executor/cuda/cuda_dnn.cc:9342] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered 2025-04-18 04:30:24.254810: E tensorflow/compiler/xla/stream_executor/cuda/cuda_fft.cc:609] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered 2025-04-18 04:30:24.254851: E tensorflow/compiler/xla/stream_executor/cuda/cuda_blas.cc:1518] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
In [1]:
# TODO: Make all necessary imports.
import os
os.environ['TF_USE_LEGACY_KERAS'] = '1'
import warnings
warnings.filterwarnings('ignore')
import tensorflow_datasets as tfds
tfds.disable_progress_bar()
import matplotlib.pyplot as plt
from tensorflow import keras
from tensorflow.keras import layers
import tensorflow as tf
import tensorflow_hub as hub
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
#matplotlib inline
%config InlineBackend.figure_format = 'retina'
import json
import numpy as np
from PIL import Image
from tensorflow.keras.models import load_model
print("TensorFlow version:", tf.__version__)
import logging
logger = tf.get_logger()
logger.setLevel(logging.ERROR)
2025-04-18 07:29:15.465254: E tensorflow/compiler/xla/stream_executor/cuda/cuda_dnn.cc:9342] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered 2025-04-18 07:29:15.465323: E tensorflow/compiler/xla/stream_executor/cuda/cuda_fft.cc:609] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered 2025-04-18 07:29:15.465366: E tensorflow/compiler/xla/stream_executor/cuda/cuda_blas.cc:1518] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
TensorFlow version: 2.14.0
Load the Dataset¶
Here you'll use tensorflow_datasets to load the Oxford Flowers 102 dataset. This dataset has 3 splits: 'train', 'test', and 'validation'. You'll also need to make sure the training data is normalized and resized to 224x224 pixels as required by the pre-trained networks.
The validation and testing sets are used to measure the model's performance on data it hasn't seen yet, but you'll still need to normalize and resize the images to the appropriate size.
In [3]:
# TODO: Load the dataset with TensorFlow Datasets.
splits = ['test[:80%]', 'test[80%:]', 'train']
#splits =['train', 'test', 'validation']
#splits = ['train[:80%]', 'train[80%:]', 'test']
datasets, dataset_info = tfds.load("oxford_flowers102", split=splits, as_supervised=True, with_info=True)
# TODO: Create a training set, a validation set and a test set.
test_dataset, train_dataset, validation_dataset = datasets
Downloading and preparing dataset 328.90 MiB (download: 328.90 MiB, generated: 331.34 MiB, total: 660.25 MiB) to /root/tensorflow_datasets/oxford_flowers102/2.1.1... Dataset oxford_flowers102 downloaded and prepared to /root/tensorflow_datasets/oxford_flowers102/2.1.1. Subsequent calls will reuse this data.
In [4]:
dataset_info
Out[4]:
tfds.core.DatasetInfo(
name='oxford_flowers102',
full_name='oxford_flowers102/2.1.1',
description="""
The Oxford Flowers 102 dataset is a consistent of 102 flower categories commonly
occurring in the United Kingdom. Each class consists of between 40 and 258
images. The images have large scale, pose and light variations. In addition,
there are categories that have large variations within the category and several
very similar categories.
The dataset is divided into a training set, a validation set and a test set. The
training set and validation set each consist of 10 images per class (totalling
1020 images each). The test set consists of the remaining 6149 images (minimum
20 per class).
Note: The dataset by default comes with a test size larger than the train size.
For more info see this
[issue](https://github.com/tensorflow/datasets/issues/3022).
""",
homepage='https://www.robots.ox.ac.uk/~vgg/data/flowers/102/',
data_dir=PosixGPath('/tmp/tmpsv6pxi_5tfds'),
file_format=tfrecord,
download_size=Unknown size,
dataset_size=331.34 MiB,
features=FeaturesDict({
'file_name': Text(shape=(), dtype=string),
'image': Image(shape=(None, None, 3), dtype=uint8),
'label': ClassLabel(shape=(), dtype=int64, num_classes=102),
}),
supervised_keys=('image', 'label'),
disable_shuffling=False,
splits={
'test': <SplitInfo num_examples=6149, num_shards=2>,
'train': <SplitInfo num_examples=1020, num_shards=1>,
'validation': <SplitInfo num_examples=1020, num_shards=1>,
},
citation="""@InProceedings{Nilsback08,
author = "Nilsback, M-E. and Zisserman, A.",
title = "Automated Flower Classification over a Large Number of Classes",
booktitle = "Proceedings of the Indian Conference on Computer Vision, Graphics and Image Processing",
year = "2008",
month = "Dec"
}""",
)
Explore the Dataset¶
In [5]:
# TODO: Get the number of examples in each set from the dataset info.
# TODO: Get the number of classes in the dataset from the dataset info.
# Get the number of examples in each set
num_train_examples = dataset_info.splits["train"].num_examples
num_test_examples = dataset_info.splits["test"].num_examples
num_validation_examples = dataset_info.splits["validation"].num_examples
# Get the number of classes
num_classes = dataset_info.features["label"].num_classes
print(f"Training examples: {num_train_examples:,}")
print(f"Testing examples: {num_test_examples:,}")
print(f"Validation examples: {num_validation_examples:,}")
print(f"Number of classes: {num_classes:,}")
Training examples: 1,020 Testing examples: 6,149 Validation examples: 1,020 Number of classes: 102
In [6]:
# TODO: Print the shape and corresponding label of 3 images in the training set.
print("Oxford Flowers 102 Training Dataset:")
for i, (image, label) in enumerate(train_dataset.take(3), start=1):
print(f"Example {i}:")
print('---------------------------------------')
print(f"\u2022 Image datatype: {image.dtype}")
print(f"\u2022 Image shape: {image.shape}")
print(f"\u2022 Label: {label.numpy()}")
print(f"\u2022 Label datatype: {label.dtype}")
print('---------------------------------------\n')
Oxford Flowers 102 Training Dataset: Example 1: --------------------------------------- • Image datatype: <dtype: 'uint8'> • Image shape: (519, 500, 3) • Label: 42 • Label datatype: <dtype: 'int64'> --------------------------------------- Example 2: --------------------------------------- • Image datatype: <dtype: 'uint8'> • Image shape: (500, 763, 3) • Label: 93 • Label datatype: <dtype: 'int64'> --------------------------------------- Example 3: --------------------------------------- • Image datatype: <dtype: 'uint8'> • Image shape: (500, 700, 3) • Label: 35 • Label datatype: <dtype: 'int64'> ---------------------------------------
Notes about the Images shapes
Inconsistent Image Shapes : The images have different shapes: (542, 500, 3), (748, 500, 3), and (500, 600, 3).Thus, resizing may be needed for uniformity in a machine learning pipeline
In [7]:
# TODO: Plot 1 image from the training set. Set the title
# of the plot to the corresponding image label.
# Get one image and its label from the training set
for image, label in train_dataset.take(1):
plt.imshow(image.numpy().squeeze())
plt.title(f"The Label of this image is: {label.numpy()}")
plt.colorbar()
plt.show()
Label Mapping¶
You'll also need to load in a mapping from label to category name. You can find this in the file label_map.json. It's a JSON object which you can read in with the json module. This will give you a dictionary mapping the integer coded labels to the actual names of the flowers.
In [14]:
with open('label_map.json', 'r') as f:
class_names = json.load(f)
class_names
Out[14]:
{'0': 'pink primrose',
'1': 'hard-leaved pocket orchid',
'2': 'canterbury bells',
'3': 'sweet pea',
'4': 'english marigold',
'5': 'tiger lily',
'6': 'moon orchid',
'7': 'bird of paradise',
'8': 'monkshood',
'9': 'globe thistle',
'10': 'snapdragon',
'11': "colt's foot",
'12': 'king protea',
'13': 'spear thistle',
'14': 'yellow iris',
'15': 'globe-flower',
'16': 'purple coneflower',
'17': 'peruvian lily',
'18': 'balloon flower',
'19': 'giant white arum lily',
'20': 'fire lily',
'21': 'pincushion flower',
'22': 'fritillary',
'23': 'red ginger',
'24': 'grape hyacinth',
'25': 'corn poppy',
'26': 'prince of wales feathers',
'27': 'stemless gentian',
'28': 'artichoke',
'29': 'sweet william',
'30': 'carnation',
'31': 'garden phlox',
'32': 'love in the mist',
'33': 'mexican aster',
'34': 'alpine sea holly',
'35': 'ruby-lipped cattleya',
'36': 'cape flower',
'37': 'great masterwort',
'38': 'siam tulip',
'39': 'lenten rose',
'40': 'barbeton daisy',
'41': 'daffodil',
'42': 'sword lily',
'43': 'poinsettia',
'44': 'bolero deep blue',
'45': 'wallflower',
'46': 'marigold',
'47': 'buttercup',
'48': 'oxeye daisy',
'49': 'common dandelion',
'50': 'petunia',
'51': 'wild pansy',
'52': 'primula',
'53': 'sunflower',
'54': 'pelargonium',
'55': 'bishop of llandaff',
'56': 'gaura',
'57': 'geranium',
'58': 'orange dahlia',
'59': 'pink-yellow dahlia?',
'60': 'cautleya spicata',
'61': 'japanese anemone',
'62': 'black-eyed susan',
'63': 'silverbush',
'64': 'californian poppy',
'65': 'osteospermum',
'66': 'spring crocus',
'67': 'bearded iris',
'68': 'windflower',
'69': 'tree poppy',
'70': 'gazania',
'71': 'azalea',
'72': 'water lily',
'73': 'rose',
'74': 'thorn apple',
'75': 'morning glory',
'76': 'passion flower',
'77': 'lotus',
'78': 'toad lily',
'79': 'anthurium',
'80': 'frangipani',
'81': 'clematis',
'82': 'hibiscus',
'83': 'columbine',
'84': 'desert-rose',
'85': 'tree mallow',
'86': 'magnolia',
'87': 'cyclamen',
'88': 'watercress',
'89': 'canna lily',
'90': 'hippeastrum',
'91': 'bee balm',
'92': 'ball moss',
'93': 'foxglove',
'94': 'bougainvillea',
'95': 'camellia',
'96': 'mallow',
'97': 'mexican petunia',
'98': 'bromelia',
'99': 'blanket flower',
'100': 'trumpet creeper',
'101': 'blackberry lily'}
In [8]:
with open('/kaggle/input/project-image-classifier/label_map.json', 'r') as f:
class_names = json.load(f)
class_names
Out[8]:
{'0': 'pink primrose',
'1': 'hard-leaved pocket orchid',
'2': 'canterbury bells',
'3': 'sweet pea',
'4': 'english marigold',
'5': 'tiger lily',
'6': 'moon orchid',
'7': 'bird of paradise',
'8': 'monkshood',
'9': 'globe thistle',
'10': 'snapdragon',
'11': "colt's foot",
'12': 'king protea',
'13': 'spear thistle',
'14': 'yellow iris',
'15': 'globe-flower',
'16': 'purple coneflower',
'17': 'peruvian lily',
'18': 'balloon flower',
'19': 'giant white arum lily',
'20': 'fire lily',
'21': 'pincushion flower',
'22': 'fritillary',
'23': 'red ginger',
'24': 'grape hyacinth',
'25': 'corn poppy',
'26': 'prince of wales feathers',
'27': 'stemless gentian',
'28': 'artichoke',
'29': 'sweet william',
'30': 'carnation',
'31': 'garden phlox',
'32': 'love in the mist',
'33': 'mexican aster',
'34': 'alpine sea holly',
'35': 'ruby-lipped cattleya',
'36': 'cape flower',
'37': 'great masterwort',
'38': 'siam tulip',
'39': 'lenten rose',
'40': 'barbeton daisy',
'41': 'daffodil',
'42': 'sword lily',
'43': 'poinsettia',
'44': 'bolero deep blue',
'45': 'wallflower',
'46': 'marigold',
'47': 'buttercup',
'48': 'oxeye daisy',
'49': 'common dandelion',
'50': 'petunia',
'51': 'wild pansy',
'52': 'primula',
'53': 'sunflower',
'54': 'pelargonium',
'55': 'bishop of llandaff',
'56': 'gaura',
'57': 'geranium',
'58': 'orange dahlia',
'59': 'pink-yellow dahlia?',
'60': 'cautleya spicata',
'61': 'japanese anemone',
'62': 'black-eyed susan',
'63': 'silverbush',
'64': 'californian poppy',
'65': 'osteospermum',
'66': 'spring crocus',
'67': 'bearded iris',
'68': 'windflower',
'69': 'tree poppy',
'70': 'gazania',
'71': 'azalea',
'72': 'water lily',
'73': 'rose',
'74': 'thorn apple',
'75': 'morning glory',
'76': 'passion flower',
'77': 'lotus',
'78': 'toad lily',
'79': 'anthurium',
'80': 'frangipani',
'81': 'clematis',
'82': 'hibiscus',
'83': 'columbine',
'84': 'desert-rose',
'85': 'tree mallow',
'86': 'magnolia',
'87': 'cyclamen',
'88': 'watercress',
'89': 'canna lily',
'90': 'hippeastrum',
'91': 'bee balm',
'92': 'ball moss',
'93': 'foxglove',
'94': 'bougainvillea',
'95': 'camellia',
'96': 'mallow',
'97': 'mexican petunia',
'98': 'bromelia',
'99': 'blanket flower',
'100': 'trumpet creeper',
'101': 'blackberry lily'}
In [9]:
# TODO: Plot 1 image from the training set. Set the title
# of the plot to the corresponding class name.
for image, label in train_dataset.take(1):
label_index = label.numpy() # Convert tensor to integer
class_name = class_names[str(label_index)] # Get class name from JSON
# Plot the image
plt.imshow(image)
plt.title(f"Class: {class_name}")
plt.colorbar()
plt.show()
Create Pipeline¶
In [10]:
# Set batch size
BATCH_SIZE = 64
In [11]:
# Define preprocessing function
def preprocess(image, label):
image = tf.image.resize(image, (224, 224)) # Resize images to 224x224
image = tf.cast(image, tf.float32) / 255.0 # Normalize pixel values to [0, 1]
return image, label
In [12]:
# TODO: Create a pipeline for each set.
train_batches = train_dataset.cache().shuffle(num_train_examples//4).map(preprocess).batch(BATCH_SIZE).cache().prefetch(1)
validation_batches = validation_dataset.cache().shuffle(num_validation_examples//4).map(preprocess).batch(BATCH_SIZE).cache().prefetch(1)
testing_batches = test_dataset.cache().shuffle(num_test_examples//4).map(preprocess).batch(BATCH_SIZE).cache().prefetch(1)
print("Pipelines created successfully!")
Pipelines created successfully!
In [13]:
for image_batch, label_batch in train_batches.take(1):
print("Images in each batch have:")
print("---------------------------------------")
print(f"\u2022 Datatype: {image_batch.dtype}")
print(f"\u2022 Shape: {image_batch.shape}")
print("---------------------------------------")
print(f"\nThis batch contains {label_batch.numpy().size} images with corresponding labels:")
print(label_batch.numpy())
print("---------------------------------------")
Images in each batch have: --------------------------------------- • Datatype: <dtype: 'float32'> • Shape: (64, 224, 224, 3) --------------------------------------- This batch contains 64 images with corresponding labels: [82 84 49 73 27 93 27 76 18 90 72 43 98 65 24 72 36 17 86 22 84 51 73 65 79 87 13 11 80 22 74 91 73 51 12 87 87 17 92 64 92 81 91 59 95 39 54 69 74 51 64 27 14 73 94 75 88 16 97 59 97 37 55 56] ---------------------------------------
In [14]:
# Take one batch
for image_batch, label_batch in train_batches.take(1):
images = image_batch.numpy().squeeze() # Convert to NumPy array
labels = label_batch.numpy() # Extract label
# Plot the first image
first_image = images[0]
first_label = labels[0]
plt.imshow(first_image)
plt.title(f"Label: {first_label}")
plt.colorbar()
plt.show()
Build and Train the Classifier¶
Now that the data is ready, it's time to build and train the classifier. You should use the MobileNet pre-trained model from TensorFlow Hub to get the image features. Build and train a new feed-forward classifier using those features.
We're going to leave this part up to you. If you want to talk through it with someone, chat with your fellow students!
Refer to the rubric for guidance on successfully completing this section. Things you'll need to do:
- Load the MobileNet pre-trained network from TensorFlow Hub.
- Define a new, untrained feed-forward network as a classifier.
- Train the classifier.
- Plot the loss and accuracy values achieved during training for the training and validation set.
- Save your trained model as a Keras model.
We've left a cell open for you below, but use as many as you need. Our advice is to break the problem up into smaller parts you can run separately. Check that each part is doing what you expect, then move on to the next. You'll likely find that as you work through each part, you'll need to go back and modify your previous code. This is totally normal!
When training make sure you're updating only the weights of the feed-forward network. You should be able to get the validation accuracy above 70% if you build everything right.
Note for Workspace users: One important tip if you're using the workspace to run your code: To avoid having your workspace disconnect during the long-running tasks in this notebook, please read in the earlier page in this lesson called Intro to GPU Workspaces about Keeping Your Session Active. You'll want to include code from the workspace_utils.py module. Also, If your model is over 1 GB when saved as a checkpoint, there might be issues with saving backups in your workspace. If your saved checkpoint is larger than 1 GB (you can open a terminal and check with ls -lh), you should reduce the size of your hidden layers and train again.
In [15]:
# URL for MobileNetV2 feature extractor pretrained on ImageNet
mobilenet_v2 ="https://tfhub.dev/google/tf2-preview/mobilenet_v2/classification/4"
# Load MobileNetV2 as a frozen feature extractor
feature_extractor = hub.KerasLayer(
mobilenet_v2,
input_shape=(224, 224, 3),
trainable=False
)
In [16]:
model = tf.keras.Sequential([feature_extractor,
tf.keras.layers.Dense(num_classes, activation='softmax')])
model.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
keras_layer (KerasLayer) (None, 1001) 3540265
dense (Dense) (None, 102) 102204
=================================================================
Total params: 3642469 (13.89 MB)
Trainable params: 102204 (399.23 KB)
Non-trainable params: 3540265 (13.51 MB)
_________________________________________________________________
In [20]:
# Compile the model
model.compile(
optimizer=tf.keras.optimizers.Adam(learning_rate=0.001),
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
#add early stopping to prevent overfitting
early_stopping= tf.keras.callbacks.EarlyStopping(
monitor='val_loss',
patience=5,
restore_best_weights=True
)
model_checkpoint = ModelCheckpoint(
'best_model.h5',
monitor='val_loss',
save_best_only=True,
mode='min',
verbose=1 # Shows when model is saved
)
In [22]:
# Set training parameters
EPOCHS = 100
# Train the model
history = model.fit(
train_batches,
validation_data=validation_batches,
epochs=EPOCHS,
callbacks=[early_stopping, model_checkpoint]
)
print("Model trained and saved successfully!")
Epoch 1/100 20/20 [==============================] - ETA: 0s - loss: 4.0857 - accuracy: 0.1675 Epoch 1: val_loss improved from inf to 3.51735, saving model to best_model.h5 20/20 [==============================] - 52s 2s/step - loss: 4.0857 - accuracy: 0.1675 - val_loss: 3.5173 - val_accuracy: 0.2245 Epoch 2/100 20/20 [==============================] - ETA: 0s - loss: 1.6821 - accuracy: 0.6293 Epoch 2: val_loss improved from 3.51735 to 2.08343, saving model to best_model.h5 20/20 [==============================] - 41s 2s/step - loss: 1.6821 - accuracy: 0.6293 - val_loss: 2.0834 - val_accuracy: 0.5265 Epoch 3/100 20/20 [==============================] - ETA: 0s - loss: 0.8095 - accuracy: 0.8577 Epoch 3: val_loss improved from 2.08343 to 1.58640, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.8095 - accuracy: 0.8577 - val_loss: 1.5864 - val_accuracy: 0.6461 Epoch 4/100 20/20 [==============================] - ETA: 0s - loss: 0.4805 - accuracy: 0.9366 Epoch 4: val_loss improved from 1.58640 to 1.38647, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.4805 - accuracy: 0.9366 - val_loss: 1.3865 - val_accuracy: 0.6784 Epoch 5/100 20/20 [==============================] - ETA: 0s - loss: 0.3197 - accuracy: 0.9675 Epoch 5: val_loss improved from 1.38647 to 1.29850, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.3197 - accuracy: 0.9675 - val_loss: 1.2985 - val_accuracy: 0.6951 Epoch 6/100 20/20 [==============================] - ETA: 0s - loss: 0.2310 - accuracy: 0.9821 Epoch 6: val_loss improved from 1.29850 to 1.24911, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.2310 - accuracy: 0.9821 - val_loss: 1.2491 - val_accuracy: 0.7039 Epoch 7/100 20/20 [==============================] - ETA: 0s - loss: 0.1757 - accuracy: 0.9951 Epoch 7: val_loss improved from 1.24911 to 1.21482, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.1757 - accuracy: 0.9951 - val_loss: 1.2148 - val_accuracy: 0.7118 Epoch 8/100 20/20 [==============================] - ETA: 0s - loss: 0.1389 - accuracy: 0.9959 Epoch 8: val_loss improved from 1.21482 to 1.19012, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.1389 - accuracy: 0.9959 - val_loss: 1.1901 - val_accuracy: 0.7127 Epoch 9/100 20/20 [==============================] - ETA: 0s - loss: 0.1129 - accuracy: 0.9984 Epoch 9: val_loss improved from 1.19012 to 1.17088, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.1129 - accuracy: 0.9984 - val_loss: 1.1709 - val_accuracy: 0.7147 Epoch 10/100 20/20 [==============================] - ETA: 0s - loss: 0.0941 - accuracy: 0.9984 Epoch 10: val_loss improved from 1.17088 to 1.15557, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0941 - accuracy: 0.9984 - val_loss: 1.1556 - val_accuracy: 0.7176 Epoch 11/100 20/20 [==============================] - ETA: 0s - loss: 0.0799 - accuracy: 0.9992 Epoch 11: val_loss improved from 1.15557 to 1.14289, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0799 - accuracy: 0.9992 - val_loss: 1.1429 - val_accuracy: 0.7225 Epoch 12/100 20/20 [==============================] - ETA: 0s - loss: 0.0690 - accuracy: 1.0000 Epoch 12: val_loss improved from 1.14289 to 1.13231, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0690 - accuracy: 1.0000 - val_loss: 1.1323 - val_accuracy: 0.7235 Epoch 13/100 20/20 [==============================] - ETA: 0s - loss: 0.0604 - accuracy: 1.0000 Epoch 13: val_loss improved from 1.13231 to 1.12331, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0604 - accuracy: 1.0000 - val_loss: 1.1233 - val_accuracy: 0.7245 Epoch 14/100 20/20 [==============================] - ETA: 0s - loss: 0.0535 - accuracy: 1.0000 Epoch 14: val_loss improved from 1.12331 to 1.11563, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0535 - accuracy: 1.0000 - val_loss: 1.1156 - val_accuracy: 0.7255 Epoch 15/100 20/20 [==============================] - ETA: 0s - loss: 0.0478 - accuracy: 1.0000 Epoch 15: val_loss improved from 1.11563 to 1.10895, saving model to best_model.h5 20/20 [==============================] - 44s 2s/step - loss: 0.0478 - accuracy: 1.0000 - val_loss: 1.1089 - val_accuracy: 0.7245 Epoch 16/100 20/20 [==============================] - ETA: 0s - loss: 0.0431 - accuracy: 1.0000 Epoch 16: val_loss improved from 1.10895 to 1.10319, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0431 - accuracy: 1.0000 - val_loss: 1.1032 - val_accuracy: 0.7245 Epoch 17/100 20/20 [==============================] - ETA: 0s - loss: 0.0390 - accuracy: 1.0000 Epoch 17: val_loss improved from 1.10319 to 1.09805, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0390 - accuracy: 1.0000 - val_loss: 1.0980 - val_accuracy: 0.7235 Epoch 18/100 20/20 [==============================] - ETA: 0s - loss: 0.0357 - accuracy: 1.0000 Epoch 18: val_loss improved from 1.09805 to 1.09366, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0357 - accuracy: 1.0000 - val_loss: 1.0937 - val_accuracy: 0.7255 Epoch 19/100 20/20 [==============================] - ETA: 0s - loss: 0.0326 - accuracy: 1.0000 Epoch 19: val_loss improved from 1.09366 to 1.08956, saving model to best_model.h5 20/20 [==============================] - 45s 2s/step - loss: 0.0326 - accuracy: 1.0000 - val_loss: 1.0896 - val_accuracy: 0.7265 Epoch 20/100 20/20 [==============================] - ETA: 0s - loss: 0.0303 - accuracy: 0.9992 Epoch 20: val_loss improved from 1.08956 to 1.08626, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0303 - accuracy: 0.9992 - val_loss: 1.0863 - val_accuracy: 0.7275 Epoch 21/100 20/20 [==============================] - ETA: 0s - loss: 0.0279 - accuracy: 0.9992 Epoch 21: val_loss improved from 1.08626 to 1.08285, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0279 - accuracy: 0.9992 - val_loss: 1.0829 - val_accuracy: 0.7275 Epoch 22/100 20/20 [==============================] - ETA: 0s - loss: 0.0264 - accuracy: 0.9992 Epoch 22: val_loss improved from 1.08285 to 1.08046, saving model to best_model.h5 20/20 [==============================] - 44s 2s/step - loss: 0.0264 - accuracy: 0.9992 - val_loss: 1.0805 - val_accuracy: 0.7275 Epoch 23/100 20/20 [==============================] - ETA: 0s - loss: 0.0246 - accuracy: 0.9992 Epoch 23: val_loss improved from 1.08046 to 1.07773, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0246 - accuracy: 0.9992 - val_loss: 1.0777 - val_accuracy: 0.7284 Epoch 24/100 20/20 [==============================] - ETA: 0s - loss: 0.0238 - accuracy: 0.9992 Epoch 24: val_loss improved from 1.07773 to 1.07552, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0238 - accuracy: 0.9992 - val_loss: 1.0755 - val_accuracy: 0.7284 Epoch 25/100 20/20 [==============================] - ETA: 0s - loss: 0.0221 - accuracy: 0.9992 Epoch 25: val_loss improved from 1.07552 to 1.07400, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0221 - accuracy: 0.9992 - val_loss: 1.0740 - val_accuracy: 0.7294 Epoch 26/100 20/20 [==============================] - ETA: 0s - loss: 0.0212 - accuracy: 0.9992 Epoch 26: val_loss improved from 1.07400 to 1.07158, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0212 - accuracy: 0.9992 - val_loss: 1.0716 - val_accuracy: 0.7294 Epoch 27/100 20/20 [==============================] - ETA: 0s - loss: 0.0196 - accuracy: 0.9992 Epoch 27: val_loss improved from 1.07158 to 1.07063, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0196 - accuracy: 0.9992 - val_loss: 1.0706 - val_accuracy: 0.7294 Epoch 28/100 20/20 [==============================] - ETA: 0s - loss: 0.0189 - accuracy: 0.9992 Epoch 28: val_loss improved from 1.07063 to 1.06856, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0189 - accuracy: 0.9992 - val_loss: 1.0686 - val_accuracy: 0.7284 Epoch 29/100 20/20 [==============================] - ETA: 0s - loss: 0.0176 - accuracy: 0.9992 Epoch 29: val_loss improved from 1.06856 to 1.06791, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0176 - accuracy: 0.9992 - val_loss: 1.0679 - val_accuracy: 0.7275 Epoch 30/100 20/20 [==============================] - ETA: 0s - loss: 0.0169 - accuracy: 0.9992 Epoch 30: val_loss improved from 1.06791 to 1.06609, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0169 - accuracy: 0.9992 - val_loss: 1.0661 - val_accuracy: 0.7275 Epoch 31/100 20/20 [==============================] - ETA: 0s - loss: 0.0159 - accuracy: 0.9992 Epoch 31: val_loss improved from 1.06609 to 1.06569, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0159 - accuracy: 0.9992 - val_loss: 1.0657 - val_accuracy: 0.7275 Epoch 32/100 20/20 [==============================] - ETA: 0s - loss: 0.0153 - accuracy: 0.9992 Epoch 32: val_loss improved from 1.06569 to 1.06408, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0153 - accuracy: 0.9992 - val_loss: 1.0641 - val_accuracy: 0.7265 Epoch 33/100 20/20 [==============================] - ETA: 0s - loss: 0.0144 - accuracy: 0.9992 Epoch 33: val_loss improved from 1.06408 to 1.06389, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0144 - accuracy: 0.9992 - val_loss: 1.0639 - val_accuracy: 0.7255 Epoch 34/100 20/20 [==============================] - ETA: 0s - loss: 0.0139 - accuracy: 0.9992 Epoch 34: val_loss improved from 1.06389 to 1.06244, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0139 - accuracy: 0.9992 - val_loss: 1.0624 - val_accuracy: 0.7255 Epoch 35/100 20/20 [==============================] - ETA: 0s - loss: 0.0130 - accuracy: 0.9992 Epoch 35: val_loss improved from 1.06244 to 1.06242, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0130 - accuracy: 0.9992 - val_loss: 1.0624 - val_accuracy: 0.7265 Epoch 36/100 20/20 [==============================] - ETA: 0s - loss: 0.0126 - accuracy: 0.9992 Epoch 36: val_loss improved from 1.06242 to 1.06110, saving model to best_model.h5 20/20 [==============================] - 44s 2s/step - loss: 0.0126 - accuracy: 0.9992 - val_loss: 1.0611 - val_accuracy: 0.7275 Epoch 37/100 20/20 [==============================] - ETA: 0s - loss: 0.0119 - accuracy: 0.9992 Epoch 37: val_loss did not improve from 1.06110 20/20 [==============================] - 42s 2s/step - loss: 0.0119 - accuracy: 0.9992 - val_loss: 1.0612 - val_accuracy: 0.7275 Epoch 38/100 20/20 [==============================] - ETA: 0s - loss: 0.0114 - accuracy: 0.9992 Epoch 38: val_loss improved from 1.06110 to 1.05999, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0114 - accuracy: 0.9992 - val_loss: 1.0600 - val_accuracy: 0.7284 Epoch 39/100 20/20 [==============================] - ETA: 0s - loss: 0.0108 - accuracy: 0.9992 Epoch 39: val_loss did not improve from 1.05999 20/20 [==============================] - 43s 2s/step - loss: 0.0108 - accuracy: 0.9992 - val_loss: 1.0602 - val_accuracy: 0.7284 Epoch 40/100 20/20 [==============================] - ETA: 0s - loss: 0.0104 - accuracy: 1.0000 Epoch 40: val_loss improved from 1.05999 to 1.05902, saving model to best_model.h5 20/20 [==============================] - 44s 2s/step - loss: 0.0104 - accuracy: 1.0000 - val_loss: 1.0590 - val_accuracy: 0.7284 Epoch 41/100 20/20 [==============================] - ETA: 0s - loss: 0.0098 - accuracy: 1.0000 Epoch 41: val_loss did not improve from 1.05902 20/20 [==============================] - 42s 2s/step - loss: 0.0098 - accuracy: 1.0000 - val_loss: 1.0593 - val_accuracy: 0.7284 Epoch 42/100 20/20 [==============================] - ETA: 0s - loss: 0.0094 - accuracy: 1.0000 Epoch 42: val_loss improved from 1.05902 to 1.05814, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0094 - accuracy: 1.0000 - val_loss: 1.0581 - val_accuracy: 0.7294 Epoch 43/100 20/20 [==============================] - ETA: 0s - loss: 0.0089 - accuracy: 1.0000 Epoch 43: val_loss did not improve from 1.05814 20/20 [==============================] - 42s 2s/step - loss: 0.0089 - accuracy: 1.0000 - val_loss: 1.0583 - val_accuracy: 0.7294 Epoch 44/100 20/20 [==============================] - ETA: 0s - loss: 0.0086 - accuracy: 1.0000 Epoch 44: val_loss improved from 1.05814 to 1.05758, saving model to best_model.h5 20/20 [==============================] - 44s 2s/step - loss: 0.0086 - accuracy: 1.0000 - val_loss: 1.0576 - val_accuracy: 0.7304 Epoch 45/100 20/20 [==============================] - ETA: 0s - loss: 0.0083 - accuracy: 1.0000 Epoch 45: val_loss improved from 1.05758 to 1.05748, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0083 - accuracy: 1.0000 - val_loss: 1.0575 - val_accuracy: 0.7304 Epoch 46/100 20/20 [==============================] - ETA: 0s - loss: 0.0080 - accuracy: 1.0000 Epoch 46: val_loss improved from 1.05748 to 1.05716, saving model to best_model.h5 20/20 [==============================] - 44s 2s/step - loss: 0.0080 - accuracy: 1.0000 - val_loss: 1.0572 - val_accuracy: 0.7304 Epoch 47/100 20/20 [==============================] - ETA: 0s - loss: 0.0077 - accuracy: 1.0000 Epoch 47: val_loss improved from 1.05716 to 1.05696, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0077 - accuracy: 1.0000 - val_loss: 1.0570 - val_accuracy: 0.7314 Epoch 48/100 20/20 [==============================] - ETA: 0s - loss: 0.0075 - accuracy: 1.0000 Epoch 48: val_loss improved from 1.05696 to 1.05678, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0075 - accuracy: 1.0000 - val_loss: 1.0568 - val_accuracy: 0.7314 Epoch 49/100 20/20 [==============================] - ETA: 0s - loss: 0.0072 - accuracy: 1.0000 Epoch 49: val_loss improved from 1.05678 to 1.05663, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0072 - accuracy: 1.0000 - val_loss: 1.0566 - val_accuracy: 0.7304 Epoch 50/100 20/20 [==============================] - ETA: 0s - loss: 0.0070 - accuracy: 1.0000 Epoch 50: val_loss improved from 1.05663 to 1.05651, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0070 - accuracy: 1.0000 - val_loss: 1.0565 - val_accuracy: 0.7304 Epoch 51/100 20/20 [==============================] - ETA: 0s - loss: 0.0068 - accuracy: 1.0000 Epoch 51: val_loss improved from 1.05651 to 1.05642, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0068 - accuracy: 1.0000 - val_loss: 1.0564 - val_accuracy: 0.7304 Epoch 52/100 20/20 [==============================] - ETA: 0s - loss: 0.0066 - accuracy: 1.0000 Epoch 52: val_loss improved from 1.05642 to 1.05636, saving model to best_model.h5 20/20 [==============================] - 41s 2s/step - loss: 0.0066 - accuracy: 1.0000 - val_loss: 1.0564 - val_accuracy: 0.7304 Epoch 53/100 20/20 [==============================] - ETA: 0s - loss: 0.0064 - accuracy: 1.0000 Epoch 53: val_loss improved from 1.05636 to 1.05632, saving model to best_model.h5 20/20 [==============================] - 43s 2s/step - loss: 0.0064 - accuracy: 1.0000 - val_loss: 1.0563 - val_accuracy: 0.7304 Epoch 54/100 20/20 [==============================] - ETA: 0s - loss: 0.0062 - accuracy: 1.0000 Epoch 54: val_loss improved from 1.05632 to 1.05630, saving model to best_model.h5 20/20 [==============================] - 42s 2s/step - loss: 0.0062 - accuracy: 1.0000 - val_loss: 1.0563 - val_accuracy: 0.7314 Epoch 55/100 20/20 [==============================] - ETA: 0s - loss: 0.0060 - accuracy: 1.0000 Epoch 55: val_loss did not improve from 1.05630 20/20 [==============================] - 42s 2s/step - loss: 0.0060 - accuracy: 1.0000 - val_loss: 1.0563 - val_accuracy: 0.7314 Epoch 56/100 20/20 [==============================] - ETA: 0s - loss: 0.0058 - accuracy: 1.0000 Epoch 56: val_loss did not improve from 1.05630 20/20 [==============================] - 42s 2s/step - loss: 0.0058 - accuracy: 1.0000 - val_loss: 1.0563 - val_accuracy: 0.7324 Epoch 57/100 20/20 [==============================] - ETA: 0s - loss: 0.0056 - accuracy: 1.0000 Epoch 57: val_loss did not improve from 1.05630 20/20 [==============================] - 43s 2s/step - loss: 0.0056 - accuracy: 1.0000 - val_loss: 1.0564 - val_accuracy: 0.7324 Epoch 58/100 20/20 [==============================] - ETA: 0s - loss: 0.0055 - accuracy: 1.0000 Epoch 58: val_loss did not improve from 1.05630 20/20 [==============================] - 44s 2s/step - loss: 0.0055 - accuracy: 1.0000 - val_loss: 1.0564 - val_accuracy: 0.7324 Epoch 59/100 20/20 [==============================] - ETA: 0s - loss: 0.0053 - accuracy: 1.0000 Epoch 59: val_loss did not improve from 1.05630 20/20 [==============================] - 43s 2s/step - loss: 0.0053 - accuracy: 1.0000 - val_loss: 1.0565 - val_accuracy: 0.7324 Model trained and saved successfully!
In [23]:
# TODO: Plot the loss and accuracy values achieved during training for the training and validation set.
# Extract loss and accuracy from history
train_loss = history.history['loss']
val_loss = history.history['val_loss']
train_acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
epochs = range(1, len(train_loss) + 1)
# Plot accuracy
plt.figure(figsize=(12, 5))
plt.subplot(1, 2, 1)
plt.plot(epochs, train_acc, 'bo-', label="Training Accuracy")
plt.plot(epochs, val_acc, 'r*-', label="Validation Accuracy")
plt.xlabel("Epochs")
plt.ylabel("Accuracy")
plt.title("Training & Validation Accuracy")
plt.legend()
# Plot loss
plt.subplot(1, 2, 2)
plt.plot(epochs, train_loss, 'bo-', label="Training Loss")
plt.plot(epochs, val_loss, 'r*-', label="Validation Loss")
plt.xlabel("Epochs")
plt.ylabel("Loss")
plt.title("Training & Validation Loss")
plt.legend()
plt.show()
Save the Model¶
Now that your network is trained, save the model so you can load it later for making inference. In the cell below save your model as a Keras model (i.e. save it as an HDF5 file).
The model is automatically saved using the following code to best_model.h5
model_checkpoint = ModelCheckpoint(
'best_model.h5',
monitor='val_loss',
save_best_only=True,
mode='min',
verbose=1 # Shows when model is saved
)
Load the Keras Model¶
Load the Keras model you saved above.
In [2]:
# Force Keras 2.x compatibility
version_fn = getattr(tf.keras, "version", None)
if version_fn and version_fn().startswith("3."):
from tf_keras.models import load_model # Use legacy Keras loader
else:
from tensorflow.keras.models import load_model # Standard loader
# Load the model
reloaded_model = tf.keras.models.load_model('best_model.h5', custom_objects = {'KerasLayer':hub.KerasLayer})
reloaded_model.summary()
2025-04-18 07:29:43.185542: E tensorflow/compiler/xla/stream_executor/cuda/cuda_driver.cc:268] failed call to cuInit: UNKNOWN ERROR (34)
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
keras_layer (KerasLayer) (None, 1001) 3540265
dense (Dense) (None, 102) 102204
=================================================================
Total params: 3642469 (13.89 MB)
Trainable params: 102204 (399.23 KB)
Non-trainable params: 3540265 (13.51 MB)
_________________________________________________________________
Testing your Network¶
It's good practice to test your trained network on test data, images the network has never seen either in training or validation. This will give you a good estimate for the model's performance on completely new images. You should be able to reach around 70% accuracy on the test set if the model has been trained well.
In [27]:
loss, accuracy = reloaded_model.evaluate(testing_batches)
print("Loss on testing set: ", loss)
print("Accuracy on testing set: ", accuracy)
77/77 [==============================] - 99s 1s/step - loss: 0.8077 - accuracy: 0.7959 Loss on testing set: 0.8077448010444641 Accuracy on testing set: 0.7958934903144836
Inference for Classification¶
Now you'll write a function that uses your trained network for inference. Write a function called predict that takes an image, a model, and then returns the top $K$ most likely class labels along with the probabilities. The function call should look like:
probs, classes = predict(image_path, model, top_k)
If top_k=5 the output of the predict function should be something like this:
probs, classes = predict(image_path, model, 5)
print(probs)
print(classes)
> [ 0.01558163 0.01541934 0.01452626 0.01443549 0.01407339]
> ['70', '3', '45', '62', '55']
Your predict function should use PIL to load the image from the given image_path. You can use the Image.open function to load the images. The Image.open() function returns an Image object. You can convert this Image object to a NumPy array by using the np.asarray() function.
The predict function will also need to handle pre-processing the input image such that it can be used by your model. We recommend you write a separate function called process_image that performs the pre-processing. You can then call the process_image function from the predict function.
Image Pre-processing¶
The process_image function should take in an image (in the form of a NumPy array) and return an image in the form of a NumPy array with shape (224, 224, 3).
First, you should convert your image into a TensorFlow Tensor and then resize it to the appropriate size using tf.image.resize.
Second, the pixel values of the input images are typically encoded as integers in the range 0-255, but the model expects the pixel values to be floats in the range 0-1. Therefore, you'll also need to normalize the pixel values.
Finally, convert your image back to a NumPy array using the .numpy() method.
In [3]:
# TODO: Create the process_image function
def process_image(image_path):
image = Image.open(image_path)
image = np.asarray(image)
image = tf.image.resize(image, (244,244))
image = tf.cast(image, tf.float32)
image /= 255
return image
To check your process_image function we have provided 4 images in the ./test_images/ folder:
- cautleya_spicata.jpg
- hard-leaved_pocket_orchid.jpg
- orange_dahlia.jpg
- wild_pansy.jpg
The code below loads one of the above images using PIL and plots the original image alongside the image produced by your process_image function. If your process_image function works, the plotted image should be the correct size.
I used Kaggle Workspace to train this model because I encountered persistent disconnection issues with the Udacity workspace. Despite reaching out to both Udacity support and my session lead, neither provided a viable solution. As a result, I explored alternative platforms and chose Kaggle for stability.
In [29]:
test_images = "/kaggle/input/project-image-classifier" # Path to the test images
In [30]:
# Loop through the files in the directory and process only .jpg files
for test_image_path in os.listdir(test_images):
# Check if the file is a .jpg file
if test_image_path.endswith('.jpg'):
fig, (ax1, ax2) = plt.subplots(figsize=(10,10), ncols=2)
# Open and display the original image
test_image = Image.open(os.path.join(test_images, test_image_path))
test_image = np.asarray(test_image)
ax1.imshow(test_image)
ax1.set_title('Original Image')
# Process the test image (you would have a process_image function)
processed_test_image = process_image(os.path.join(test_images, test_image_path))
ax2.imshow(processed_test_image)
ax2.set_title('Processed Image')
plt.tight_layout()
plt.show()
Once you can get images in the correct format, it's time to write the predict function for making inference with your model.
Inference¶
Remember, the predict function should take an image, a model, and then returns the top $K$ most likely class labels along with the probabilities. The function call should look like:
probs, classes = predict(image_path, model, top_k)
If top_k=5 the output of the predict function should be something like this:
probs, classes = predict(image_path, model, 5)
print(probs)
print(classes)
> [ 0.01558163 0.01541934 0.01452626 0.01443549 0.01407339]
> ['70', '3', '45', '62', '55']
Your predict function should use PIL to load the image from the given image_path. You can use the Image.open function to load the images. The Image.open() function returns an Image object. You can convert this Image object to a NumPy array by using the np.asarray() function.
Note: The image returned by the process_image function is a NumPy array with shape (224, 224, 3) but the model expects the input images to be of shape (1, 224, 224, 3). This extra dimension represents the batch size. We suggest you use the np.expand_dims() function to add the extra dimension.
Sanity Check¶
It's always good to check the predictions made by your model to make sure they are correct. To check your predictions we have provided 4 images in the ./test_images/ folder:
- cautleya_spicata.jpg
- hard-leaved_pocket_orchid.jpg
- orange_dahlia.jpg
- wild_pansy.jpg
In the cell below use matplotlib to plot the input image alongside the probabilities for the top 5 classes predicted by your model. Plot the probabilities as a bar graph. The plot should look like this:
You can convert from the class integer labels to actual flower names using class_names.
In [16]:
#test_images='test_images'
In [7]:
def process_image(image_path):
image = Image.open(image_path)
image = np.asarray(image)
image = tf.image.resize(image, (224, 224))
image = tf.cast(image, tf.float32)
image /= 255
return image
In [8]:
def predict(image_path, model, top_k=5):
processed_img = process_image(image_path)
# Add batch dimension
processed_img = np.expand_dims(processed_img, axis=0)
prob_pred = model.predict(processed_img)
probs, classes = tf.math.top_k(prob_pred, k=top_k)
probs = probs.numpy().tolist()[0]
classes = classes.numpy().tolist()[0]
return probs, classes
In [9]:
# Loop through the files in the directory and process only .jpg files
for test_image_path in os.listdir(test_images):
if test_image_path.lower().endswith('.jpg'): # Check if file is a .jpg (case insensitive)
full_path = os.path.join(test_images, test_image_path)
print(full_path)
probs, classes = predict(full_path, reloaded_model, 5)
print(probs, classes)
test_images/cautleya_spicata.jpg 1/1 [==============================] - 0s 472ms/step [0.995582640171051, 0.0007080113282427192, 0.0006947956862859428, 0.0005628669168800116, 0.00021319069492165] [60, 17, 23, 72, 14] test_images/orange_dahlia.jpg 1/1 [==============================] - 0s 153ms/step [0.9902433753013611, 0.004151088185608387, 0.0017485009739175439, 0.0013808460207656026, 0.0005885145510546863] [58, 4, 65, 40, 99] test_images/wild_pansy.jpg 1/1 [==============================] - 0s 78ms/step [0.9993693232536316, 0.0001576405338710174, 0.0001234916999237612, 9.114976273849607e-05, 4.960298974765465e-05] [51, 47, 65, 64, 80] test_images/hard-leaved_pocket_orchid.jpg 1/1 [==============================] - 0s 159ms/step [0.9980308413505554, 0.0007707187905907631, 0.0001885326491901651, 0.00015477377746719867, 0.00013757504348177463] [1, 76, 79, 90, 6]
In [17]:
def plot_prop(test_image_path, processed_image, top_k_class_names, top_k_probs):
"""
Plots the input image alongside its predicted class probabilities.
Args:
test_image_path (str): Path to the original image file
processed_image (array): Processed image array (after resizing/normalization)
top_k_class_names (list): List of top K class names
top_k_probs (list): List of corresponding probabilities
"""
fig, (ax1, ax2) = plt.subplots(figsize=(12, 6), ncols=2)
# Plot image
ax1.imshow(processed_image)
ax1.axis('off')
# Format title by removing file extension and underscores
title = os.path.basename(test_image_path).replace('_', ' ').replace(".jpg", "")
ax1.set_title(title, fontsize=12, pad=10)
# Plot probabilities
ax2.barh(np.arange(len(top_k_probs)), top_k_probs, color='skyblue')
ax2.set_aspect(0.1)
ax2.set_yticks(np.arange(len(top_k_probs)))
ax2.set_yticklabels(top_k_class_names, size=10)
# Add probability text on bars
for i, prob in enumerate(top_k_probs):
ax2.text(prob + 0.02, i, f"{prob:.2f}", va='center', fontsize=10)
ax2.set_title('Class Probabilities', fontsize=12, pad=10)
ax2.set_xlim(0, 1.1)
ax2.grid(axis='x', linestyle='--', alpha=0.7)
plt.tight_layout()
plt.show()
In [21]:
for test_image_path in os.listdir(test_images):
# Only process .jpg files (case insensitive)
if not test_image_path.lower().endswith('.jpg'):
continue
try:
# Open and process the image
test_image = Image.open(os.path.join(test_images, test_image_path))
# Get predictions (only 2 values)
top_k_probs, top_k_classes = predict(
os.path.join(test_images, test_image_path),
reloaded_model,
5
)
# Map class indices to names
top_k_class_names = [class_names[str(label)].title() for label in top_k_classes]
# Plot the results (using the original image instead of processed_image)
plot_prop(test_image_path, np.array(test_image), top_k_class_names, top_k_probs)
except Exception as e:
print(f"Error processing {test_image_path}: {str(e)}")
continue
1/1 [==============================] - 0s 74ms/step
1/1 [==============================] - 0s 43ms/step
1/1 [==============================] - 0s 59ms/step
1/1 [==============================] - 0s 61ms/step
In [19]:
for test_image_path in os.listdir(test_images):
if not test_image_path.lower().endswith(('.jpg', '.jpeg', '.png')):
continue
full_path = os.path.join(test_images, test_image_path)
try:
# Get predictions (only 2 values)
top_k_probs, top_k_classes = predict(full_path, reloaded_model, 5)
# Load original image for display
display_image = Image.open(full_path)
display_image = np.array(display_image)
# Map class indices to names
top_k_class_names = [class_names[str(cls)].title() for cls in top_k_classes]
plot_prop(test_image_path, display_image, top_k_class_names, top_k_probs)
except Exception as e:
print(f"Error processing {test_image_path}: {str(e)}")
continue
1/1 [==============================] - 0s 60ms/step
1/1 [==============================] - 0s 86ms/step
1/1 [==============================] - 0s 54ms/step
1/1 [==============================] - 0s 52ms/step
Part two - Test predict.py¶
In [20]:
!python predict.py --input "test_images/wild_pansy.jpg" --top_k 5
2025-04-18 07:34:44.388307: E tensorflow/compiler/xla/stream_executor/cuda/cuda_dnn.cc:9342] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered 2025-04-18 07:34:44.388365: E tensorflow/compiler/xla/stream_executor/cuda/cuda_fft.cc:609] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered 2025-04-18 07:34:44.388430: E tensorflow/compiler/xla/stream_executor/cuda/cuda_blas.cc:1518] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered Starting Prediction... 2025-04-18 07:34:46.861148: E tensorflow/compiler/xla/stream_executor/cuda/cuda_driver.cc:268] failed call to cuInit: UNKNOWN ERROR (34) 1/1 [==============================] - 0s 385ms/step Results: ============================== Image: test_images/wild_pansy.jpg • wild pansy: 99.94% --------------------------- • buttercup: 0.02% --------------------------- • osteospermum: 0.01% --------------------------- • californian poppy: 0.01% --------------------------- • frangipani: 0.00% ---------------------------
Save as HTML¶
In [8]:
# Convert the current notebook to HTML
!jupyter nbconvert --to html project-image-classifier-project-two.ipynb
[NbConvertApp] Converting notebook project-image-classifier-project-two.ipynb to html [NbConvertApp] WARNING | Alternative text is missing on 14 image(s). [NbConvertApp] Writing 13454766 bytes to project-image-classifier-project-two.html
In [ ]: